We used two base hardware platforms: a Hewlett-Packard LPr (450-MHz Pentium II, 192M bytes dynamic random-access memory [DRAM]; Symbios Logic drive controller with two 8G-byte Seagate drives; Intel 10/100 Ethernet network interface card [NIC]) and a Compaq ProLiant 1600 (twin 450-MHz Pentium IIs, Compaq Smart Array/DH, four 8.3G-byte Seagate drives, Compaq Netelligent 10/100 Ethernet NICs). Also used on both platforms was a Mylex 1100 64M-byte UltraWide SCSI-3 Array Controller connected to an Andataco 8-SCSI drive array with 8.3G-byte Seagate drives. We installed the Mylex 1100 Array controller with suggested factory defaults and allocated 50% of cache to reads and 50% to writes.
Clients included 12 PCs and notebooks running Windows 95/98, Windows NT 4.0 Workstation, RedHat and Caldera Linux, and FreeBSD Unix. Each server was tested out-of-the-box using only supplied components with Windows NT 4.0 (described elsewhere) and Caldera OpenLinux 2.2, which uses the Linux 2.2 kernel. Each platform was retested using the Mylex 1100 RAID controller. We used Internet Security Systems' SecureScanner 5.6 to test the servers for security holes on the default platform, then the default platform with vendor-suggested security fixes. We used WilsonWindowWare's WinBatch to launch a batch file on Windows platforms, a bash script to trigger Linux and the FreeBSD clients into a series of file reads and writes to the server, and then timed their execution. Averages were taken after 10 test iterations; data fell within 5% of the averages shown. The clients were free of any other nondefault processes running at the time. The following processes were running on NT and Linux server platforms: Domain Name System (with upstream authority); DHCP (no leases); Web service (Internet Information Server and Apache, respectively); the user interface (Windows graphical user interface and XFree86) and File Transfer Protocol (FTP) service. Server Message Block access to Linux was provided by Samba for all clients connected. We also performed a simple FTP put test, which wrote a 100M-byte file for two and four identical senders (HP 4440 Pavilions with 333-MHz Celeron CPU and 64M bytes DRAM) on to the tested server. We tracked CPU utilization using the X CPU application on OpenLinux and PerfMon on NT Server. RELATED LINKSBack to the main review
